Artificial Intelligence/Machine Learning, ASA(ALT), Direct to Phase II
Dynamic Generative Large Language Model for Continuous Situational Awareness
Objective
The proposed SBIR topic aims to advance the capabilities of large language models by addressing critical challenges and enhancing functionalities relevant to military applications, particularly within the U.S. Army.
Description
This topic will accept Direct to Phase II submission only. DP2 proposals are accepted for a cost up to $2 million for an 18-month period of performance.
This project focuses on developing methodologies to detect and mitigate bias in model outputs, ensuring the generation of fair and unbiased information. It also seeks innovative solutions to identify and correct hallucinations (false information generation) to bolster the reliability of LLMs. Furthermore, the integration of multimodal inputs and outputs will be explored to broaden the application scope of LLMs beyond text, facilitating their use in analyzing diverse data types such as images and videos.
Enhancements in text summarization are also targeted to efficiently condense large volumes of information into actionable intelligence. The ability to rapidly train, fine-tune, and/or augment with external data source LLMs in specialized focus areas such as Acquisition, Intelligence, Operations, and Logistics, enabling tailored applications that meet specific Army needs is desired. Lastly, this project should assist in identifying metrics for quantifying LLM performance to easily discern which trained models are best for a task.
Phase I
This topic is only accepting DP2 proposals for a cost up to $2 million for an 18-month period of performance.
Proposers interested in submitting a DP2 proposal must provide documentation to substantiate that the scientific and technical merit and feasibility equivalent to a Phase I project has been met. Documentation can include data, reports, specific measurements, success criteria of a prototype, etc.
The focus of this SBIR topic is Dynamic Generative LLM for Continuous Situational Awareness technology is proficiency in a wide range of language tasks, including text generation, translation, summarization and question answering. These technologies demonstrate that foundational knowledge and methods already exist, thus not requiring a feasibility study. Models like GPT-3 have demonstrated impressive capabilities in understanding context, generating coherent text, and even engaging in rudimentary forms of reasoning and problem-solving.
However, challenges remain, such as mitigating biases, improving contextual understanding, and ensuring ethical usage. Despite these challenges, LLM technology has reached a level of maturity where it’s being actively applied in various industries, from customer service and content generation to healthcare and finance, albeit with ongoing refinement and development. These foundational technologies can be leveraged for this SBIR topic and adapted for Department of Defense and Army use cases without requiring a feasibility study.
Phase II
During DP2, firms should research and develop an improved data labeling approach, design and develop initial prototype, and test prototype on representative/surrogate data sets.
Phase III
- LLM COAA technology is encompassed by the broader space of Situational Awareness. Situational Awareness solutions play a crucial role in enabling real-time decision-making, risk mitigation, and operational efficiency in complex environments.
- Development of COAs in military operations is traditionally a time-consuming process. COA-GPT, is a novel algorithm employing LLMs for rapid and efficient generation of military plans that address planning discrepancies and capitalize on emergent windows of opportunity.
- Military/Defense applications of open-source data and publicly available information utilized in COAA LLMs for U.S. Special Operation Forces include:
- Data fusion for Advanced Situational Awareness
- Geospatial Pattern Analysis & Self-Location in GPS-Denied Environments (celestial imagery to enable self-location without GPS for ground operators using ATAK)
- Indicators and Warnings: examine tactical, operational, and strategic variables across the operational environment
- Dual-use applications include:
- Autonomous systems (Unmanned Aerial Systems /drones)
- Smart cities/homes
- Internet of Things sensors for home and business
- Predictive analytics for climate and energy applications
- Disaster management
Submission Information
For more information, and to submit your full proposal package, visit the DSIP Portal.
SBIR|STTR Help Desk: usarmy.sbirsttr@army.mil
References:
- https://www.sciencedirect.com/science/article/abs/pii/S1041608023000195
- https://doi.org/10.1007/s11704-024-40231-1
- KEYWORDS: LLM, Large Language Model, AI/ML, GenAI, Generative AI, Hallucinations, multimodal
Objective
The proposed SBIR topic aims to advance the capabilities of large language models by addressing critical challenges and enhancing functionalities relevant to military applications, particularly within the U.S. Army.
Description
This topic will accept Direct to Phase II submission only. DP2 proposals are accepted for a cost up to $2 million for an 18-month period of performance.
This project focuses on developing methodologies to detect and mitigate bias in model outputs, ensuring the generation of fair and unbiased information. It also seeks innovative solutions to identify and correct hallucinations (false information generation) to bolster the reliability of LLMs. Furthermore, the integration of multimodal inputs and outputs will be explored to broaden the application scope of LLMs beyond text, facilitating their use in analyzing diverse data types such as images and videos.
Enhancements in text summarization are also targeted to efficiently condense large volumes of information into actionable intelligence. The ability to rapidly train, fine-tune, and/or augment with external data source LLMs in specialized focus areas such as Acquisition, Intelligence, Operations, and Logistics, enabling tailored applications that meet specific Army needs is desired. Lastly, this project should assist in identifying metrics for quantifying LLM performance to easily discern which trained models are best for a task.
Phase I
This topic is only accepting DP2 proposals for a cost up to $2 million for an 18-month period of performance.
Proposers interested in submitting a DP2 proposal must provide documentation to substantiate that the scientific and technical merit and feasibility equivalent to a Phase I project has been met. Documentation can include data, reports, specific measurements, success criteria of a prototype, etc.
The focus of this SBIR topic is Dynamic Generative LLM for Continuous Situational Awareness technology is proficiency in a wide range of language tasks, including text generation, translation, summarization and question answering. These technologies demonstrate that foundational knowledge and methods already exist, thus not requiring a feasibility study. Models like GPT-3 have demonstrated impressive capabilities in understanding context, generating coherent text, and even engaging in rudimentary forms of reasoning and problem-solving.
However, challenges remain, such as mitigating biases, improving contextual understanding, and ensuring ethical usage. Despite these challenges, LLM technology has reached a level of maturity where it’s being actively applied in various industries, from customer service and content generation to healthcare and finance, albeit with ongoing refinement and development. These foundational technologies can be leveraged for this SBIR topic and adapted for Department of Defense and Army use cases without requiring a feasibility study.
Phase II
During DP2, firms should research and develop an improved data labeling approach, design and develop initial prototype, and test prototype on representative/surrogate data sets.
Phase III
- LLM COAA technology is encompassed by the broader space of Situational Awareness. Situational Awareness solutions play a crucial role in enabling real-time decision-making, risk mitigation, and operational efficiency in complex environments.
- Development of COAs in military operations is traditionally a time-consuming process. COA-GPT, is a novel algorithm employing LLMs for rapid and efficient generation of military plans that address planning discrepancies and capitalize on emergent windows of opportunity.
- Military/Defense applications of open-source data and publicly available information utilized in COAA LLMs for U.S. Special Operation Forces include:
- Data fusion for Advanced Situational Awareness
- Geospatial Pattern Analysis & Self-Location in GPS-Denied Environments (celestial imagery to enable self-location without GPS for ground operators using ATAK)
- Indicators and Warnings: examine tactical, operational, and strategic variables across the operational environment
- Dual-use applications include:
- Autonomous systems (Unmanned Aerial Systems /drones)
- Smart cities/homes
- Internet of Things sensors for home and business
- Predictive analytics for climate and energy applications
- Disaster management
Submission Information
For more information, and to submit your full proposal package, visit the DSIP Portal.
SBIR|STTR Help Desk: usarmy.sbirsttr@army.mil
References:
- https://www.sciencedirect.com/science/article/abs/pii/S1041608023000195
- https://doi.org/10.1007/s11704-024-40231-1
- KEYWORDS: LLM, Large Language Model, AI/ML, GenAI, Generative AI, Hallucinations, multimodal